DecoEvo: Score-Decoupled Co-Evolution of Solver and Rubric-Generator Skills in Text Space
Not on the map. Why?
Abstract (arXiv)
Text-space optimization adapts large language models (LLMs) by editing external natural-language artifacts rather than model weights, so the optimized artifacts remain inspectable and the model can be treated as a black box. However, most existing text-space methods keep evaluation fixed. On open-ended tasks, this can become a bottleneck: once the solver improves on the criteria a rubric measures, omitted dimensions remain invisible to the optimization signal. Simply evolving the rubric is also unreliable when updates are selected by the current solver's score, because apparent progress can come from making the rubric easier to satisfy. We introduce DecoEvo (Decoupled Co-Evolution), which co-evolves a solver skill and a rubric-generator skill under decoupled objectives without using gold rubrics during optimization. The solver skill is updated using criterion-level feedback, while the rubric-generator skill is revised through complementary audits of requirement coverage and response discrimination that are independent of aggregate solver score. This separation focuses generator updates on newly exposed solver weaknesses, reducing repeated emphasis on criteria the solver already satisfies. Under each benchmark's official evaluation, DecoEvo outperforms all compared methods across five benchmarks and three LLM backbones, yielding 2.8--5.0% relative gains over SkillOpt in the five-benchmark average.
Key facts. Cited by 2 papers in the map (data through 14 Oct 2026). Builds on 25 of its 32 references that are in the map. First posted on arXiv 28 Jul 2026.
Builds on
25 of 32 references.
References · 25
- 2022Training language models to follow instructions with human feedbackcited by 15,971
- 2025Qwen3 Technical Reportcited by 8,820
- 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
- 2024GPT-4o System Cardcited by 4,588
- 2023G-Eval: NLG Evaluation using GPT-4 with Better Human Alignmentcited by 2,276
- 2022Scaling Laws for Reward Model Overoptimizationcited by 1,095
- 2023Large Language Models as Optimizerscited by 864
- 2023Prometheus: Inducing Fine-grained Evaluation Capability in Language Modelscited by 575
- 2025HealthBench: Evaluating Large Language Models Towards Improved Human Healthcited by 429
- 2023JudgeLM: Fine-tuned Large Language Models are Scalable Judgescited by 367
- 2026SkillOpt: Executive Strategy for Self-Evolving Agent Skillscited by 148
- 2025OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignmentcited by 132
13 more
- 2025DR Tulu: Reinforcement Learning with Evolving Rubrics for Deep Researchcited by 114
- 2025WritingBench: A Comprehensive Benchmark for Generative Writingcited by 106
- 2025ResearchQA: Evaluating Scholarly Question Answering at Scale Across 75 Fields with Survey-Mined Questions and Rubricscited by 49
- 2025Online Rubrics Elicitation from Pairwise Comparisonscited by 46
- 2025AMO-Bench: Large Language Models Still Struggle in High School Math Competitionscited by 38
- 2025LLMEval-Med: A Real-world Clinical Benchmark for Medical LLMs with Physician Validationcited by 34
- 2026Open Rubric System: Scaling Reinforcement Learning with Pairwise Adaptive Rubriccited by 27
- 2026EvoRubrics: Dynamic Rubrics as Rewards via Adversarial Co-Evolution for LLM Reinforcement Learningcited by 14
- 2026EvoLM: Self-Evolving Language Models through Co-Evolved Discriminative Rubricscited by 13
- 2026EvoRubric: Self-Evolving Rubric-Driven RL for Open-Ended Generationcited by 11
- 2026General365: Benchmarking General Reasoning in Large Language Models Across Diverse and Challenging Taskscited by 6
- 2026From Proprietary to Open-Source: Bridging the Distribution Gap via Multi-Agent Protocol Distillation in Agentic Searchcited by 5
- 2026ARCO: Adaptive Rubrics with Co-Evolution for Multi-Step LLM-Based Agentscited by 4
Built on by
Citing papers per year: 2026: 2.
2 papers.